Three Claude Code plugins hit the Show HN front page in the same week. That's unusual — the plugin ecosystem has been quiet for months, and suddenly there are three worth evaluating. Here's the breakdown: two installs, one reaction post worth reading before you build, and one vertical connector to skip unless your client roster includes law firms.
Version Sentinel: stop hallucinated dependency versions before they ship
The supply-chain attack problem with AI-assisted coding isn't that Claude writes bad code. It's that Claude installs whatever package version its training data remembers — which may be stale, yanked, or in one recent case, actively credential-stealing at a million monthly downloads. Version Sentinel solves this at the PreToolUse hook level, before the install executes.
When Claude tries to add or bump a dependency — npm install [email protected], pip install requests, cargo add tokio — the hook fires first. If no fresh registry lookup is on record for that package-and-version pair, the tool call exits with code 2:
BLOCKED: version-sentinel. Package: lodash (npm). Version: 4.17.21. No fresh version check on record.
Claude then runs a web search, invokes /vs-record to log the result, and retries. The first install per package adds about 15 seconds. After that, the record is cached for the session.
Version 0.1 covers five ecosystems: npm, PyPI, Cargo, NuGet, and Go modules. The install is one line:
/plugin marketplace add https://github.com/KSEGIT/Version-Sentinel.git
We wrote up the broader case for this pattern — and why post-hoc CI auditors aren't sufficient — in our earlier piece on guardrails inside Claude Code. The short version: if a bad version ships, you want to know exactly when it was added and what registry lookup Claude cited. Version Sentinel gives you that paper trail. Install it.
Archik: YAML-to-architecture diagrams without the diagramming tax
Archik takes YAML service definitions and generates architecture diagrams. If you've tried to keep a Mermaid diagram synced with real infrastructure — a Next.js app on Vercel, Supabase for auth and database, Cloudflare at the edge, Make.com wiring automations together — you know the maintenance cost is high enough that most teams let the docs rot within a month.
The Archik approach: define your services once in YAML, let the plugin regenerate the diagram on every Claude session. The diagram lives in the repo next to the code. When the stack changes, update the YAML and re-run. No drag-and-drop canvas, no subscription.
We tested it against the Get Indiana stack: three core services, two external integrations. The diagram was accurate and readable on the first pass. The YAML schema is simple enough that a junior developer can maintain it without guidance — if they understand the actual infrastructure, they can write the YAML.
The real value isn't the diagram itself. It's that the diagram stays current because the cost of updating it is low. Install this if your team has any infrastructure worth documenting, which is most teams.
The ticketing automation: a build guide, not an off-the-shelf install
A Show HN post this week demoed Claude as a ticket-creation layer: paste a bug report or feature request into Claude, get a formatted ticket ready to push to Linear or Jira. It's a clean idea. The implementation details are where it gets complicated.
We built an equivalent for a Tuscan client earlier this year — chat → ticket → assignee routing via Make.com and Claude — and the full build notes, including the two failure modes we hit, are in our writeup on wiring Claude into a ticketing system. The short version: Claude structures tickets well, but routing logic — which project, which assignee, what priority — needs explicit rules or it guesses wrong at the edges.
The Show HN implementation is closer to a proof of concept than a production pattern. Don't treat it as a plugin install. It's a 2-3 hour build with Make.com as the glue layer. Read it as inspiration, then build your own version with the routing rules your client's workflow actually requires.
The one to skip: Clio MCP
Clio MCP wires Claude into Clio, the practice management software used by a large portion of small law firms. The technical work is solid — the connector exposes matters, contacts, and time entries to Claude's context window. For a law-firm-focused agency, that's a real productivity tool.
For a general digital agency, it's a dead end. The connector only pays off if you're billing hours into Clio and want Claude to pull matter context automatically. Unless you have law firm clients, skip it.
The broader point: as the plugin marketplace grows, vertical connectors will outnumber horizontal tools. Most won't apply to your stack. The evaluation question isn't "is this technically good?" — it usually is. The question is whether the tool's scope matches the work you actually do. Clio MCP is a well-built answer to a question most agencies aren't asking.
Plugins as a toolchain moat
The same Claude model runs in every session. What differentiates agency use of it is the toolchain around it: which plugins are installed, what's in CLAUDE.md, how work moves between Make.com and the agent loop. Version Sentinel and Archik both build toward that moat — one enforces supply-chain hygiene on every dependency decision, the other keeps infrastructure documentation alive without a maintenance burden. Neither is a feature you can copy by reading about it. They're defaults you have to actually install and run.
Install Version Sentinel and Archik today. Treat the ticketing automation as a build reference. Skip Clio unless law firm work is on the table.
